Archive
Tag: Research
-
Multi-Teacher Distillation Is Now the Standard Recipe for Frontier AI Models
MOPD (multi-teacher on-policy distillation) has replaced single-pipeline RLHF as the go-to post-training recipe for frontier models in 2026. Here’s what changed and…
-
Axiom Math Is Betting That Provable AI Beats Guessing AI
Axiom scored 12/12 on the Putnam exam and 99% on a code verification benchmark. Their thesis: AI can’t truly scale reasoning without…
-
GPT-5 Reproduced a Physicist’s Best Paper in 30 Minutes, Then Derived New Quantum Gravity Results
OpenAI’s GPT-5 reproduced peer-reviewed physics research in minutes and then generated 110 pages of novel quantum gravity calculations. Here’s what it means…
-
AI Benchmarks Are Broken. Open-World Evaluations Are the Fix.
A new research collaboration called CRUX is testing AI agents on messy, real-world tasks like publishing iOS apps. Here’s why that matters…